Aliasing Artefact Suppression in Compressed Sensing MRI for Random Phase-Encode Undersampling

Goal: Random phase-encode undersampling of Cartesian k-space trajectories is widely implemented in compressed sensing (CS) MRI. However, its one-dimensional (1-D) randomness inherently introduces large coherent aliasing artefacts along the undersampled direction in the reconstruction and, thus, degrades the image quality. This paper proposes a novel reconstruction scheme to reduce the 1-D undersampling-induced aliasing artefacts. Methods: The proposed reconstruction progress is separated into two steps in our new algorithm. In step one, we transfer the original two-dimensional (2-D) image reconstruction into a parallel 1-D signal reconstruction procedure, which takes advantage of the superior incoherence property in the phase direction. In step two, using the new k-space data obtained from the 1-D reconstructions, we implement a follow-up 2-D CS reconstruction to produce a better solution, which exploits the inherent correlations between the adjacent lines of 1-D reconstructed signals. Results: We evaluated the performance on various cases of typical MR images, including cardiac cine, brain, foot, and angiogram at the reduction factor up to 10 and compared the results with the conventional CS method. Experiments using the proposed method demonstrated faithful reconstruction of the MR images. Conclusion: Compared with conventional method, the new method achieves more accurate reconstruction results with 2-5 dB gain in peak SNR and higher structural similarity index. Significance: The proposed method improves image quality of the reconstructions and suppresses the coherent artefacts introduced by the random phase-encode undersampling.